Adversarial reverse mapping of equilibrated condensed-phase molecular structures

Hydrogen-induced failure of TiNi based alloy with coarse-grained and ultrafine-grained structure

Coarse-grained (CG) models of biomolecules have been widely used in protein/ribonucleic acid (RNA) three-dimensional structure prediction, docking, drug design, and molecular simulations due to their superiority in computational efficiency. Most of these applications strongly depend on the reasonable estimation of solvation free energy, which requires the accurate calculation of solvent accessible surface area (SASA). Although algorithms for SASA calculations with all-atom protein and RNA structures have been well-established, accurately estimating the SASA based on CG structures is extremely challenging. In this work, we developed a deep learning-based SASA estimator (DeepCGSA), which can provide almost perfect SASA estimation based on CG structures of protein and RNA molecules. Extensive testing analysis showed that for three types of widely used CG protein models, including the Cα-based, Cα-Cβ, and Martini models, the correlation coefficients between the predicted values and the reference values can be as high as 0.95-0.99, which perform dramatically better than available methods. In addition, the new method can be used for CG RNA structures and unfolded protein structures with much improved accuracy. We anticipate that DeepCGSA will be highly useful in the protein/RNA structure prediction, drug design, and other applications, in which accurate estimations of SASA for CG biomolecular structures are critically important.

A robust gray-box modeling methodology for variable-speed direct-expansion systems with limited training data

Removed due to plagiarism The original research was published as: MODELING OF ULTRASONIC PROPAGATION IN A COARSE GRAIN STRUCTURE <https://apps.webofknowledge.com/full_record.do?product=UA&search_mode=GeneralSearch&qid=2&SID=Z1pJDVNJsjgeYkWdsdi&page=2&doc=15> By: Jenson, F.; Fortuna, T.; Doudet, L. Edited by: Thompson, DO; Chimenti, DE Conference: 35th Annual Review of Progress in Quantitative Nondestructive Evaluation Location: Univ Chicago, Chicago, IL Date: JUL 20-25, 2008 REVIEW OF PROGRESS IN QUANTITATIVE NONDESTRUCTIVE EVALUATION, VOLS 28A AND 28B Book Series:AIP Conference Proceedings Volume: 1096 Pages: 1201-1208 Published: 2009

The MArtini Database (MAD - https://mad.ibcp.fr) is a web server designed for the sharing of structures and topologies of molecules parametrized with the Martini coarse-grained (CG) force field. MAD can also convert atomistic structures into CG structures and prepare complex systems (including proteins, lipids, etc.) for molecular dynamics (MD) simulations at the CG level. It is dedicated to the generation of input files for Martini 3, the most recent version of this popular CG force field. Specifically, the MAD server currently includes tools to submit or retrieve CG models of a wide range of molecules (lipids, carbohydrates, nanoparticles, etc.), transform atomistic protein structures into CG structures and topologies, with fine control on the process and assemble biomolecules into large systems, and deliver all files necessary to start simulations in the GROMACS MD engine.

Deep learning algorithms are incredibly powerful and have achieved impressive results in various classification tasks. However, one of their limitations is their dependence on large amounts of training data. When there is limited training data available, the standard approach is to increase the dataset size by using data augmentation and training a neural network on the expanded dataset. This method can be effective, but it requires significant computational resources and may not always be feasible. In our work, we proposed a new approach to address the problem of limited training data. Instead of relying solely on increasing the dataset size, we train Generative Adversarial Networks (GANs) to generate distributions of individual categories. The classification of the unknown element is then performed using distributions generated by trained GAN networks. We proposed four methods that compare the unknown element with the elements generated by trained GAN networks and establish estimates of the conditional probabilities of the unknown element belonging to individual categories. These conditional probabilities are then used for the classification of the unknown element into individual categories. This approach enables us to make informed decisions and achieve accurate classification results even when dealing with limited training data.

To obtain actual conditions of infrastructure assets and manage them more efficiently, extensive research efforts have been placed on structural health monitoring (SHM), especially those using data-driven methods. Recently, deep learning becomes a research hotspot in many application areas, including the SHM domain. Their performance largely relies on the quality and quantity of the training data, obtained either experimentally or numerically. Due to the time and expense restraints, field or laboratory test data are normally limited by the variation of structural conditions, while the quality of numerical simulation data is subjective to experts' modelling skills. Therefore, the actual performance of deep learning algorithms with limited training data needs to be studied, and the alternative ways to generate more training data need to be developed. In this work, we develop a new one-Dimensional Convolutional Neural Network (1D-CNN) for structural condition identification. A laboratory case study is conducted to evaluate the performance of the algorithm. A steel Warren truss bridge structure is constructed and instrumented with accelerometers and impact hammer. The vibration tests under seven different scenarios are conducted, and each scenario has five repeated test data. The algorithm is trained with different quantities of training data (from one test data to four test data for each scenario). The results show that condition identification results become reliable with at least three repeated test data. To overcome the challenge of limited monitoring data, we propose the potential application of Generative Adversarial Networks (GANs) to generate more reliable training data.

The hydrogen effect on the inelastic properties and plastic strain development after electrolythical hydrogenation in physiological solution was investigated. This effect virtually results in a failure under torsion of Ti49.1Ni50.9 (atom per cent) alloy specimens with coarse-grained (CG) and submicrocrystalline (SMC) structures. It is shown that hydrogen embrittlement (HE) phenomenon occurs irrespective of the grain size in the studied specimens at approximately equal strain values. However, compared to the specimens with CG structure, those with SMC structure accumulate two to three times more hydrogen for the same hydrogenation time. It is found that hydrogen has a much smaller effect on the inelastic properties of specimens with SMC structure as compared to those with CG structure.

Molecular dynamics simulations based on coarse-grained (CG) models are used to accelerate conformational dynamics of biomolecules and other chemical systems with reduced computational costs. CG models achieve this by discarding atomic information necessary for downstream structural analysis. Recovering the atomic detail from CG structures, i.e., backmapping, remains a fundamental challenge in multiscale modeling, especially for proteins and complex biomolecular assemblies. Recent machine learning methods have shown promise in reconstructing atomistic details, but most approaches only target simple systems at a small scale. In particular, conventional backmapping pipelines often fail to preserve stereochemistry, induce high-energy configurations, or require extensive minimization. Here we present CGBack, a backmapping framework that employs a denoising diffusion probabilistic model to reconstruct all-atom molecular structures from CG representations. CGBack consists of backmapping and refinement procedures and accurately recovers atomic coordinates across diverse protein systems from small to large scales. We show that CGBack is capable of accurately backmapping both single-chain and multichain molecular systems, including densely packed intrinsically disordered proteins in condensates. These results suggest that CGBack can be a powerful tool for multiscale molecular simulation pipelines. We anticipate that CGBack will enable more efficient workflows for protein modeling, as well as for other biomolecules across different CG models.

A major bottleneck for building statistical spoken dialogue systems for new domains and applications is the need for large amounts of training data. To address this problem, we adopt the multi-dimensional approach to dialogue management and evaluate its potential for transfer learning. Specifically, we exploit pre-trained task-independent policies to speed up training for an extended task-specific action set, in which the single summary action for requesting a slot is replaced by multiple slot-specific request actions. Policy optimisation and evaluation experiments using an agenda-based user simulator show that with limited training data, much better performance levels can be achieved when using the proposed multi-dimensional adaptation method. We confirm this improvement in a crowd-sourced human user evaluation of our spoken dialogue system, comparing partially trained policies. The multi-dimensional system (with adaptation on limited training data in the target scenario) outperforms the one-dimensional baseline (without adaptation on the same amount of training data) by 7% perceived success rate.

This study proposes a novel deep neural network (DNN) based approach to language identification (LID) for the NIST 2015 Language Recognition (LRE) i-Vector Machine Learning Challenge. State-of-the-art DNN based LID systems utilize large amounts of labeled training data. The 2015 LRE i-Vector Machine Learning Challenge limits the access to only ready-to-use i-Vectors for LID system training and testing. This poses unique challenges in designing DNN based LID systems, since optimized front-ends and network architectures can no longer be used. We propose to use the training i-Vectors to train an initial DNN for LID. Next, we present a novel strategy to use this initial DNN to estimate out-of-set language labels from the development data. The final DNN for LID is trained using the original training data, and the estimated out-of-set language data. We show that augmenting the training set with out-of-set labels leads to significant improvement in the LID performance. Our approach obtains very competitive costs (defined by NIST) of 26.56, and 25.98 respectively, on the progress and evaluation subsets of the challenge. Since the amount of training data is very limited (300 i-Vectors per language), this study outlines a successful recipe for DNN based LID using very limited resources.

Most studies in statistical or machine learning based authorship attribution focus on two or a few authors. This leads to an overestimation of the importance of the features extracted from the training data and found to be discriminating for these small sets of authors. Most studies also use sizes of training data that are unrealistic for situations in which stylometry is applied (e.g., forensics), and thereby overestimate the accuracy of their approach in these situations. A more realistic interpretation of the task is as an authorship verification problem that we approximate by pooling data from many different authors as negative examples. In this paper, we show, on the basis of a new corpus with 145 authors, what the effect is of many authors on feature selection and learning, and show robustness of a memory-based learning approach in doing authorship attribution and verification with many authors and limited training data when compared to eager learning methods such as SVMs and maximum entropy learning.

The purpose of this work was to study the hot-corrosion behavior of dissimilar weldments of AISI 304 stainless steel and AISI 1015 low carbon steel assembled with consumable-309L stainless steel wires at 750 °C in static air with 2 mg/cm^2 mixture of various ratios of NaCl-Na_2SO_4 deposits. It was found that the HAZ of weldments revealed the coarse grain structure and the around 304 HAZ. Moreover, the carbon migrated from the substrate of low carbon steel of the higher chromium weld metal side and formed chromium carbide adjacent to the interface between the low carbon steel HAZ and weld metal. The results of hot corrosion showed that both 304 and 309L appeared the subscale attack while low carbon steel revealed the uniform corrosion. The depth of attack increased with increasing exposure duration of time. The 309L weld metal had the smallest metal loss because of its high chromium content. Due to sensitized reaction the 304 HAZ had the highest metal loss and the total depth of attack. The galvanic effect to the low carbon steel resulted in a serious corrosion with sufficient molten salts at low carbon steel HAZ/WM interface.

Production forecasting is vital in the oil and gas sector, empowering engineers with insights for effective reservoir management. This paper introduces the concept of Transfer Learning as a powerful tool in the domain of machine-learning-assisted production forecasting that accounts of 3D spatial distributions of three geological properties, namely porosity, permeability, and saturation, two completion parameters, namely hydraulic fracture height and length, and production constraint. Transfer learning efficiently leverages knowledge from one problem to improve generalization on another, especially when data is scarce and computational resources are limited. To demonstrate the utility of transfer learning, we evaluate two scenarios of transfer learning. The first transfer learning scenario demonstrates the generalization of the forecasting to cases with variable hydraulic fracture spacing using limited training data. The second transfer learning scenario demonstrates the generalization of the forecasting to cases with variable natural fracture spacing and natural fracture permeability using limited training data. Source dataset contained 2000 realizations, while the target dataset contained 20, 40, 60, 80, 100, 250, 500, or 1000 realizations to represent the scenarios of limited training-data availability. The study confirms the benefits of transfer learning when the training dataset size is small (generally less than 100 training realizations); however, under large training dataset size (around 500 or more training realizations), transfer learning is not needed. For the first scenario involving variable hydraulic fracture spacing, the use of transfer learning ensured that the source model can be trained on target dataset with 80 realizations for gas rate forecasting at an accuracy of 25% in terms of MAPE, and with 500 realizations for condensate rate forecasting at an accuracy of 12% in terms of MAPE. Similarly, for the second scenario involving variable natural fracture spacing and natural fracture permeability, the use of transfer learning ensured that the source model can be trained on target dataset with 250 realizations for gas rate forecasting at an accuracy of 24% in terms of MAPE, and with 500 realizations for condensate rate forecasting at an accuracy of 23% in terms of MAPE. This illustrates the potential of transfer learning in improving forecasting models with limited data using a well pre-trained model and enhanced hyperparameter tuning of the transfer learning model. For cases with 500 or more training realizations, transfer learning severely underperforms as compared to training a conventional machine-learning model from scratch. The paper explores two cases of transfer learning.

Rapidly developing location acquisition technologies provide a powerful tool for understanding and predicting human mobility in cities, which is very significant for urban planning, traffic regulation, and emergency management. However, with the existing methodologies, it is still difficult to accurately predict millions of peoples’ mobility in a large urban area such as Tokyo, Shanghai, and Hong Kong, especially when collected data used for model training are often limited to a small portion of the total population. Obviously, human activities in city are closely linked with point-of-interest (POI) information, which can reflect the semantic meaning of human mobility. This motivates us to fuse human mobility data and city POI data to improve the prediction performance with limited training data, but current fusion technologies can hardly handle these two heterogeneous data. Therefore, we propose a unique POI-embedding mechanism, that aggregates the regional POIs by categories to generate an artificial POI-image for each urban grid and enriches each trajectory snippet to a four-dimensional tensor in an analogous manner to a short video. Then, we design a deep learning architecture combining CNN with LSTM to simultaneously capture both the spatiotemporal and geographical information from the enriched trajectories. Furthermore, transfer learning is employed to transfer mobility knowledge from one city to another, so that we can fully utilize other cities’ data to train a stronger model for the target city with only limited data available. Finally, we achieve satisfactory performance of human mobility prediction at the citywide level using a limited amount of trajectories as training data, which has been validated over five urban areas of different types and scales.